Evaluation of the University of California Diabetes Prevention Program (UC DPP) Initiative

The UC DPP Coordinating Center was established by UCOP/UC Health to support system-wide DPP activities. The UC DPP Coordinating Center assists UC campuses with all aspects of program delivery, including DPP coach training by UCLA-based certified master trainers and data management for DPP cohorts. Centralizing some activities, such as coach training, provides efficiency of scale (e.g., each individual campus is not required to spend time and money to train master trainers). The Coordinating Center leads monthly group calls with all UC campuses and individual calls with each campus as needed. Each UC campus is also registered with the National DPP (each campus has their own unique CDC organization number in order to be eligible for CDC recognition).

The UC system, like many other academic institutions, provides and manages health insurance benefits for many of its employees. The UC DPP Coordinating Center conducts outreach to at-risk UC health insurance beneficiaries with documented prediabetes (i.e., prediabetes diagnosed as part of usual care). Trained UCOP staff use UC EHR and claims data algorithms to identify at-risk participants with documented prediabetes. Personalized initiation letters with UC DPP logos are mailed from the Coordinating Center to overweight/obese UC beneficiaries with documented prediabetes. These prediabetes awareness letters include information about the importance of diabetes prevention and the UC DPP Initiative, including contact information for the local campus DPP. The letters clarify that there is no out-of-pocket cost for the UC DPP. UC DPP cohorts are led by UC staff who complete DPP coach training and have experience delivering campus-based wellness programs. DPP eligibility criteria include 1) age > 18 years, 2) body mass index (BMI) > 25 kg/m² (> 23 for Asian adults) and 3) history of prediabetes, gestational diabetes, or a “high risk” CDC Prediabetes Risk Test. UC funding is provided to campuses for each additional DPP cohort and is not contingent on meeting National DPP metrics (e.g., 5% weight loss).

To assess the effectiveness of the UC DPP Initiative, we will examine mean percent weight change from baseline at 12 month follow-up and 24 month follow-up within and between three groups of overweight/obese UC beneficiaries. Group 1 includes overweight/obese beneficiaries with documented prediabetes who receive UC DPP Initiative prediabetes awareness letters and enroll in DPP (i.e., receive “full treatment”). Group 2 includes overweight/obese beneficiaries with documented prediabetes who receive UC DPP Initiative prediabetes awareness letters, but do not enroll in DPP. Group 3 includes overweight/obese UC beneficiaries without documented prediabetes who are not included in the UC DPP Initiative (do not receive prediabetes awareness letters and do not enroll in DPP). Group 3 serves as a control group to account for secular trends and/or concurrent programs that may affect weight change outcomes among UC beneficiaries (such as weight management services any overweight/obese UC beneficiaries may receive as part of routine care). The primary outcome will be % weight change at 12-month follow-up within those who received a “full intervention dose” (i.e., Group 1) and secondary outcomes will compare mean % weight change between UC beneficiaries (Groups 1 vs. 2 and Group 1 + 2 vs. Group 3).

To measure mean percent weight change, we will use EHR weights, which are collected similarly across UC health center visits, and DPP session weights for UC beneficiaries who enroll in DPP. Baseline covariate differences between those with and without attrition will be examined and we will then adjust for any covariates found to be significantly different (p < 0.05) between attrition groups in order to satisfy the Missing at Random assumptions of the mixed effects models. Because attrition could be a consequence of our primary outcome (i.e. weight) our missing data may be Missing Not at Random. In this instance we will utilize pattern mixture modeling to impute missing valuies. We will conduct within-subject analysis estimates of UC DPP Initiative effectiveness among those who received a “full intervention dose” (i.e., Group 1 prediabetes awareness letters and enrolled in DPP). For this primary analysis, we will use a mixed effects linear regression model to compare percent weight change within Group 1 subjects 12 and 24 months before DPP enrollment (pre period) and 12 and 24 months after enrollment (post period). Models for the trajectory of percent weight change will be specified with a random intercept for person with nested random slope for time from enrollment measured on a continuum. Average within-person between time-period (post vs. pre-enrollment) differences in the trajectory for % weight change will be modeled using a fixed effect interaction term of time-by-time-period to test for between time-period differences in the rate of % weight change. These models will be fit using a random intercept for person and random slope for time-period (post vs pre). A fixed effect of time-period will be used to estimate the average between time-period difference in % weight change adjusting for the corresponding length of time over which the change was observed. For both models, exploratory post-hoc analyses will also consider additional subject level covariates (age, gender, race, ethnicity, number of co-morbidities, number of DPP sessions attended, education level and years employed at the institution), as well as DPP site. Interactions of the difference over time (rates or mean levels) by these subject characteristics or site will also be explored to evaluate whether there is a differential benefit of the program for certain groups of beneficiaries (i.e., heterogeneity in treatment effect) and across sites. In the presence of statistically significant (p < 0.05) interactions, the models will be stratified so as to allow for potential heterogeneity in the covariates as well.

We will also compare outcomes between UC beneficiaries in Groups 1 vs. 2, a between-subject comparison, to estimate DPP effectiveness while holding the effect of the prediabetes awareness campaign constant across groups (both Group 1 and 2 receive prediabetes outreach letters). The mixed effects models mentioned above will be modified to examine between group differences in within-person change by incorporating an interaction term with Group membership. Similarly, we will compare outcomes in Groups 1 + 2 vs. Group 3, a between-subject comparison akin to an intent-to-treat (ITT) analytic approach since outcomes are being assessed among all UC beneficiaries with documented prediabetes eligible to receive the UC DPP Initiative in comparison to similarly overweight/obese individuals not exposed to the program. We will use propensity score matching to identify comparable overweight/obese beneficiaries with (Groups 1 + 2) and without (Group 3) documented prediabetes. The propensity score model will include baseline age, gender, race/ethnicity, BMI, and medical co-morbidities. Matching will be accomplished through use of the nearest-neighbor algorithm. Between group balance in the propensity scores and covariates will additionally be examined. While Groups 1 and 2 have an easily defined baseline for differentiating between pre- and post- diabetes awareness (i.e., enrollment or awareness letter date), Group 3 will have their baseline selected as corresponding to 12–18 months (or 24–30 months) prior to their most recent EHR entry.

For our primary outcome of % weight change at 12-month follow-up within subjects, we will use 5% weight change to estimate power based on the 2002 DPP randomized controlled trial [4] as well as National DPP recognition standards [22]. With a proposed sample size of N = 50 using a two-sided alpha level of 0.05, we would demonstrate 80% power to detect a paired difference of at least 3.3% and have > 98% power to detect our target weight change of 5%. For our outcome of weight change at 12-month follow-up between subjects, we conservatively assume that the average % weight loss for those in Group 2 (prediabetes notified) and Group 3 (usual care) will be 2 and 0% respectively. These values will yield about a 3% difference in weight loss between our comparison groups. Using an independent samples t-test as a simplification of the mixed effects analysis plan with a two-sided alpha of 0.05, we will have 90% power to detect our assumed between group difference of 3% with the proposed sample sizes of 50 (Group 1; 10 per site) and 500 (Group 2; 100 per site). Using the same assumptions, the proposed sample sizes of N = 550 Group A + B and N = 500 Group 3 would yield 99% power to detect the assumed 3% difference in the percentage weight change between the groups. We would additionally have 80% power to minimally detect a between group difference of 2.5 and 1.1% with these sample sizes. All sample sizes account for an anticipated 20% loss to follow-up [7]. We will not be powered to detect differences in rates of incident type 2 diabetes, but we will examine trends over time using validated EHR and claims data algorithms [23, 24]. Metformin is not specifically highlighted in the UC DPP Initiative, but since some beneficiaries with prediabetes may also seek out this option, [4, 25] we will also examine rates of metformin uptake using prescription data available in the EHR.

To assess reach, we will report the proportion of eligible UC beneficiaries who engage in DPP (attend > 9 and > 16 sessions) and their representativeness of UC beneficiaries overall. Using DPP enrollment and participation data from each campus, we will assess the proportion of eligible participants who enroll in DPP, as well as rates of participation and engagement according to CDC standards (e.g., average number of sessions attended, proportion who completed > 9 and > 16 sessions, proportion with documented physical activity minutes, etc.). We will assess the representativeness of those who enrolled and participated in DPP as compared to the population overall, as well as the subset who were contacted for enrollment but chose not to enroll. We will also report exclusions, participation rates, drop-outs and representativeness within campuses and across UC.

We will examine the effectiveness of the DPP program (% weight change and DPP participation) within and between subjects. We will evaluate the consistency of outcomes across sites and key subgroups (e.g., race and ethnicity). We will also construct a logistic regression model to evaluate factors that predict whether subjects that receive the prediabetes letter will enroll in DPP. This is equivalent to the comparison of patient level factors between Group 1 and Group 2. First, we will use bivariate logistic regression models to evaluate subject level characteristics (e.g., age, gender, race, ethnicity, BMI, change in weight between the pre time point and baseline, income level) that predict the outcome. Next, we will construct a multivariable logistic regression model using a Least Absolute Shrinkage and Selection Operator (LASSO), [26] to identify the combination of factors that predict whether eligible subjects enroll in DPP (Group 1 vs. Group 2).

Relevant RE-AIM constructs will be used as an initial organizing framework for data coding and analysis. We will use matrix analysis methods [27] for rapid turn-around of the results, including aggregated site profiles, to share with UC leadership. In-depth analysis of the qualitative data will be conducted using ATLAS.ti. Initially, a top-level codebook will be developed for the baseline interviews based on the semi-structured interview guide, [28] Using a constant comparison analytic approach, this codebook will be elaborated upon based on emergent themes, and will be adjusted as each round of interviews is reviewed. Interviews will be compared within each campus, across campuses, across roles, and over time. Additional sources of qualitative data (i.e., field notes from site visits) will also be included in the data set and will be coded separately and in relation to the interview data.

This study was funded by NIH/NIDDK (1R01DK124503–01) and approved by the UCLA IRB (20–000357-AM-00001).